Practical Expansion and Risk Governance Evaluation of China’s Intelligent Trial Assistance System
Published 19 May 2026
Sarah Xuan
On April 28, 2026, the Shanghai High People’s Court released “An All-in-One Guide: The Whole-Process Intelligent Trial Assistance System”, introducing the new generation of the intelligent trial assistance framework in Shanghai courts. Relying on the unified case-handling and office platform of courts nationwide, this system integrates application scenarios such as “digital assistance for case handling, digital assistance for supervision, and digital assistance for public convenience”. It embeds intelligent assistance into the entire lifecycle of cases—including case filing, trial, execution, and supervision—and combines technologies such as artificial intelligence and agents with the trial experience and adjudication rules accumulated by Shanghai courts to provide support for judges’ case handling, trial supervision, and litigation services.
This system also serves as a local paradigm for the digital transformation and standardized application of judicial artificial intelligence in the People’s Courts. The Opinions of the Supreme People’s Court on Regulating and Strengthening the Judicial Application of Artificial Intelligence explicitly state that artificial intelligence must strictly adhere to its positioning as a trial assistant, and auxiliary outcomes may only serve as references for trials, supervision, and administration. Furthermore, the 2026 Work Report of the Supreme People’s Court emphasized that the research and development of artificial intelligence-assisted trial systems must proceed steadily and prudently, and the subject of judicial responsibility can only be the judge. It is precisely under this institutional framework that the Shanghai Whole-Process Intelligent Trial Assistance System represents a concrete exploration of embedding artificial intelligence into trial workflows to improve trial quality and litigation convenience.
Taking the Whole-Process Intelligent Trial Assistance System of Shanghai courts as an entry point, this article examines the practical expansion of artificial intelligence-assisted trials in courts nationwide, and further analyzes its institutional value, potential risks, and governance boundaries.
I. From the “System 206” to the Whole-Process Intelligent Trial Assistance System Shanghai courts commenced their exploration of judicial artificial intelligence relatively early. In 2017, Shanghai developed the Intelligent Assistant System for Criminal Cases, known as “System 206”. Through functions such as evidentiary standard guidelines, evidentiary verification, flaw prompts, and courtroom trial assistance, it served the reform of the trial-centered criminal litigation system, enhanced the quality of criminal case handling, and prevented wrongful convictions. In 2019, this system entered courtroom trials, enabling the retrieval of evidence and screening for flaws based on voice commands, thereby driving the transition of artificial intelligence from back-end assistance to active courtroom support.
The Whole-Process Intelligent Trial Assistance System released in 2026 represents a systematic upgrade of these early explorations. Compared to “System 206”, which primarily focused on criminal evidence review and courtroom trial assistance, the new system expands capabilities such as evidentiary verification, rule prompts, reference to similar cases, and risk early-warning to a wider array of case types, procedural nodes, and litigation participants. This marks the shift of Shanghai courts’ intelligent trial exploration from single-scenario piloting to whole-process system integration.
This upgrade is built upon the digital reform practices of Shanghai courts in recent years. Since 2023, Shanghai courts have established multiple application segments centered around “digital assistance for case handling, digital assistance for supervision, digital assistance for public convenience, digital assistance for governance, and digital assistance for government affairs”, and have conducted intelligent prompting and risk identification in scenarios such as litigant death or enterprise deregulation notifications, false litigation screening, and document risk early-warning. The Whole-Process Intelligent Trial Assistance System is precisely an integrated and systematic achievement developed on the foundation of these practices.
II. Practices of Intelligent Auxiliary Trial Systems in Courts Nationwide In recent years, courts in Beijing, Shenzhen, Zhejiang, Jiangsu, Chongqing, and other regions have explored the use of intelligent systems to assist in case trials across various levels, causes of action, and procedural stages, with each demonstrating distinct practical emphases.
Beijing courts constructed the “Digital Intelligent Beijing Law” (Shuzhi Beifa) intelligent application platform. Centered on knowledge retrieval, case file review analysis, document assistance, and trial quality and efficiency management, it has formed four categories of knowledge services: “searching, analyzing, writing, and verifying”, and has embedded large language model capabilities into stages such as case analysis, similar case retrieval, and document generation. Shenzhen courts emphasize whole-process empowerment; their artificial intelligence-assisted trial system deconstructs the process from case filing to case closure into 85 nodes, sets up modules such as intelligent case filing review, intelligent case file review, intelligent courtroom trial, and intelligent legal documents, and ensures that the process can be corrected and traced entirely through mechanisms like verification, confirmation, and prompts.
Zhejiang courts pioneered the intelligent adjudication of categorized cases. The “Phoenix Intelligent Adjudication” (Fenghuang Zhisheng) Version 2.0 primarily serves financial dispute cases, assisting in case assignment, scheduling, service of process, archiving, summary of disputed issues, calculation of adjudication amounts, and drafting of legal documents. It interfaces with the financial comprehensive service platform to form online channels for case filing, trial, and execution. The Hangzhou Internet Court has also explored full-process intelligence from case filing to adjudication in cases such as financial loan disputes. The “Future Judge’s Assistant” of Suzhou Intermediate People’s Court in Jiangsu Province focuses on intelligent case file review, case fact Q&A, document generation, and quality control error correction. The “Yi Shen” platform of Chongqing courts is based on the deep application of electronic case files, achieving functions such as automatic data backfilling, similar case pushing, intelligent legal documents, and case correlation.
Overall, the practice of intelligent auxiliary trials in courts nationwide has generally formed four tracks: Shanghai and Shenzhen focus on whole-process assistance; Zhejiang and the Hangzhou Internet Court focus on the intelligent trial of categorized cases; Beijing and Suzhou focus on large language model-assisted case file review and document generation; and Chongqing focuses on the deep application of electronic case files. The common trend is that judicial artificial intelligence is moving from voice transcription, template documents, and simple retrieval toward a comprehensive application stage deeply integrated with electronic case files, workflow management, knowledge graphs, large language models, and trial supervision.
III. Institutional Value of Intelligent Assistance Systems First and foremost, the primary value of intelligent assistance systems lies in improving trial efficiency. Faced with a long-standing high volume of cases, judges must complete case file reviews, fact summarization, evidence examination, application of law, courtroom trial organization, and document drafting within a limited timeframe. Through material recognition, element extraction, summary of disputed issues, and assisted document generation, the intelligent system shoulders repetitive and administrative labor, enabling judges to devote more energy to factual determination, rule application, and value balancing. Shenzhen courts deconstructed the trial workflow into 85 nodes and assisted in a large volume of case filings and document generation during the pilot phase. The application of Suzhou’s “Future Judge’s Assistant” in intelligent case file review, document generation, and quality control error correction also demonstrates that artificial intelligence is reshaping traditional methods of reviewing files and producing documents.
Secondly, the deeper value of intelligent assistance systems lies in promoting the uniformity of adjudication standards. In the past, achieving similar judgments for similar cases primarily relied on judges’ proactive retrieval, ex-post case reviews, and guidance from higher courts. In contrast, intelligent assistance systems shift reference to similar cases, key rules, discretionary factors, and risk alerts forward into the case-handling process, enabling judges to obtain rule-based support before a judgment is formed. By transforming trial experience, adjudication rules, and case-handling essentials into a professional knowledge base, the Shanghai system essentially converts scattered, experiential, and tacit judicial knowledge into reusable, callable, and traceable rule resources.
Furthermore, intelligent assistance systems have altered the methods of case quality control. Traditional case evaluations are mostly retrospective, sample-based, and static, making it difficult to detect procedural flaws, factual omissions, and risks in the application of law in a timely manner. Through workflow node verification, document quality control, abnormal data identification, and risk early-warning, the intelligent system embeds supervision directly into the case-handling process. The large language model-based intelligent trial supervision of Beijing’s “Digital Intelligent Beijing Law” has been deployed to automatically screen for abnormal clues and alert users to risks in trial procedures and the application of law, reflecting a shift in trial supervision from ex-post error correction to process-oriented governance.
Moreover, for litigants, intelligent assistance systems lower the threshold for litigation through material prompts, procedural guidance, online interactions, and intelligent consultation. Following the integration of Zhejiang’s “Phoenix Intelligent Adjudication” with the financial comprehensive service platform, online channels for case filing, trial, and execution have been established. The Hangzhou Internet Court has explored full-process intelligence from case filing to the generation of judgment documents in specific cases such as financial loan contract disputes, breaking through the time and spatial constraints of traditional courtroom trials via an asynchronous trial mode.
Looking further, intelligent assistance systems have also enhanced judicial governance capabilities. When case elements, dispute types, industry risks, and litigation behaviors are structurally identified, courts can discover false litigation, batch disputes, hidden industry hazards, and weak links in governance from a massive volume of cases. Consequently, the intelligent trial assistance system is evolving from a case-handling support tool into an important digital infrastructure for the People’s Courts to participate in social governance.
IV. Risk Governance of Intelligent Auxiliary Trials The more deeply an intelligent assistance system penetrates trial workflows, the greater the need to simultaneously establish institutional constraints. The crux of judicial artificial intelligence lies in how these outputs are utilized, reviewed, and subjected to accountability.
First, algorithmic bias must be prevented. Judicial adjudication involves factual determination, admissibility of evidence, application of law, and value judgment. If training data contains historical biases, or if models develop inappropriate inclinations in similar case recommendations, risk identification, and document generation, the system may amplify existing issues. Therefore, normalized evaluations should be carried out for key functions such as factual element extraction, similar case recommendations, risk early-warning, and document generation, focusing on testing accuracy, stability, bias risks, false positives/negatives, and explainability. The processes of significant model updates and invocations in major cases must be tracked to ensure ex-post traceability and reviewability.
Second, automation bias and over-reliance must be prevented. As systems continuously provide summaries of disputed issues, adjudication paths, and preliminary document drafts, judges may tend to accept system conclusions under the pressure of efficiency, or even become subject to de facto algorithmic constraints. In response, the “auxiliary positioning” must be translated into explicit rules of authority: the system may flag risks, push similar cases, and summarize facts, but the final judgment on case outcomes must be completed independently by the judge. System outputs that exert a substantive impact on case disposition must undergo human review, with confirmations, corrections, or explanations for deviation recorded where necessary. The reasoning of a judgment must be independently articulated by the judge using legal language, and system conclusions must not replace judicial reasoning.
Third, the structure of liability must be clarified. Intelligent assistance systems are constructed with the joint participation of courts, technology enterprises, data governance teams, model training teams, and operations and maintenance (O&M) personnel. If a system erroneously extracts facts, omits evidence, mismatches similar cases, or generates faulty documents, liability cannot simply be attributed to the notion that “the judge is ultimately responsible”. While judges are responsible for adjudication conclusions, system design defects, data quality issues, and technical O&M risks should also be integrated into the scope of auditing, record-filing, and accountability. Courts must internally distinguish among judges’ review responsibilities, trial management responsibilities, technical O&M responsibilities, and suppliers’ security responsibilities, thereby avoiding situations where ambiguous liability leads judges to fear using the systems or where technical errors remain undetected.
In addition, data security and privacy protection must serve as fundamental constraints. Courts possess a vast amount of sensitive materials, including identity, property, commercial secrets, personal privacy, juvenile information, and criminal case records. Once case data is utilized for model training, cross-platform collaboration, and system O&M, risks of data leakage, unauthorized access, model inversion, and improper third-party contact inevitably rise. To address this, mechanisms for data classification, access control, desensitization processing, log tracking, and third-party security review should be established, safeguarding judicial security and the rights and interests of litigants through refined data governance.
The institutional goal of intelligent auxiliary trials is to preserve the subjectivity of judicial judgment while simultaneously enhancing efficiency, unifying adjudication, and strengthening supervision. Intelligent systems excel at discovering similarities, whereas judicial adjudication must identify differences. A mature intelligent trial assistance system should help judges determine where cases are identical, where they differ, and where explanations are required, rather than compressing the space for individualized justice through technical outputs.
Conclusion The boundaries of judicial artificial intelligence cannot be determined solely by technical capability itself, but must be jointly delineated by the operation laws of judicial power, procedural justice, and the ethics of responsibility. The standard for evaluating a whole-process intelligent trial assistance system should not merely be its technical advancement or the scale of its efficiency gains, but rather whether it achieves a higher degree of understandable, supervisable, and accountable judicial justice. Therefore, the ultimate goal of digital court construction is to render justice more prudent, transparent, and trustworthy under the support of technology.
This system also serves as a local paradigm for the digital transformation and standardized application of judicial artificial intelligence in the People’s Courts. The Opinions of the Supreme People’s Court on Regulating and Strengthening the Judicial Application of Artificial Intelligence explicitly state that artificial intelligence must strictly adhere to its positioning as a trial assistant, and auxiliary outcomes may only serve as references for trials, supervision, and administration. Furthermore, the 2026 Work Report of the Supreme People’s Court emphasized that the research and development of artificial intelligence-assisted trial systems must proceed steadily and prudently, and the subject of judicial responsibility can only be the judge. It is precisely under this institutional framework that the Shanghai Whole-Process Intelligent Trial Assistance System represents a concrete exploration of embedding artificial intelligence into trial workflows to improve trial quality and litigation convenience.
Taking the Whole-Process Intelligent Trial Assistance System of Shanghai courts as an entry point, this article examines the practical expansion of artificial intelligence-assisted trials in courts nationwide, and further analyzes its institutional value, potential risks, and governance boundaries.
I. From the “System 206” to the Whole-Process Intelligent Trial Assistance System Shanghai courts commenced their exploration of judicial artificial intelligence relatively early. In 2017, Shanghai developed the Intelligent Assistant System for Criminal Cases, known as “System 206”. Through functions such as evidentiary standard guidelines, evidentiary verification, flaw prompts, and courtroom trial assistance, it served the reform of the trial-centered criminal litigation system, enhanced the quality of criminal case handling, and prevented wrongful convictions. In 2019, this system entered courtroom trials, enabling the retrieval of evidence and screening for flaws based on voice commands, thereby driving the transition of artificial intelligence from back-end assistance to active courtroom support.
The Whole-Process Intelligent Trial Assistance System released in 2026 represents a systematic upgrade of these early explorations. Compared to “System 206”, which primarily focused on criminal evidence review and courtroom trial assistance, the new system expands capabilities such as evidentiary verification, rule prompts, reference to similar cases, and risk early-warning to a wider array of case types, procedural nodes, and litigation participants. This marks the shift of Shanghai courts’ intelligent trial exploration from single-scenario piloting to whole-process system integration.
This upgrade is built upon the digital reform practices of Shanghai courts in recent years. Since 2023, Shanghai courts have established multiple application segments centered around “digital assistance for case handling, digital assistance for supervision, digital assistance for public convenience, digital assistance for governance, and digital assistance for government affairs”, and have conducted intelligent prompting and risk identification in scenarios such as litigant death or enterprise deregulation notifications, false litigation screening, and document risk early-warning. The Whole-Process Intelligent Trial Assistance System is precisely an integrated and systematic achievement developed on the foundation of these practices.
II. Practices of Intelligent Auxiliary Trial Systems in Courts Nationwide In recent years, courts in Beijing, Shenzhen, Zhejiang, Jiangsu, Chongqing, and other regions have explored the use of intelligent systems to assist in case trials across various levels, causes of action, and procedural stages, with each demonstrating distinct practical emphases.
Beijing courts constructed the “Digital Intelligent Beijing Law” (Shuzhi Beifa) intelligent application platform. Centered on knowledge retrieval, case file review analysis, document assistance, and trial quality and efficiency management, it has formed four categories of knowledge services: “searching, analyzing, writing, and verifying”, and has embedded large language model capabilities into stages such as case analysis, similar case retrieval, and document generation. Shenzhen courts emphasize whole-process empowerment; their artificial intelligence-assisted trial system deconstructs the process from case filing to case closure into 85 nodes, sets up modules such as intelligent case filing review, intelligent case file review, intelligent courtroom trial, and intelligent legal documents, and ensures that the process can be corrected and traced entirely through mechanisms like verification, confirmation, and prompts.
Zhejiang courts pioneered the intelligent adjudication of categorized cases. The “Phoenix Intelligent Adjudication” (Fenghuang Zhisheng) Version 2.0 primarily serves financial dispute cases, assisting in case assignment, scheduling, service of process, archiving, summary of disputed issues, calculation of adjudication amounts, and drafting of legal documents. It interfaces with the financial comprehensive service platform to form online channels for case filing, trial, and execution. The Hangzhou Internet Court has also explored full-process intelligence from case filing to adjudication in cases such as financial loan disputes. The “Future Judge’s Assistant” of Suzhou Intermediate People’s Court in Jiangsu Province focuses on intelligent case file review, case fact Q&A, document generation, and quality control error correction. The “Yi Shen” platform of Chongqing courts is based on the deep application of electronic case files, achieving functions such as automatic data backfilling, similar case pushing, intelligent legal documents, and case correlation.
Overall, the practice of intelligent auxiliary trials in courts nationwide has generally formed four tracks: Shanghai and Shenzhen focus on whole-process assistance; Zhejiang and the Hangzhou Internet Court focus on the intelligent trial of categorized cases; Beijing and Suzhou focus on large language model-assisted case file review and document generation; and Chongqing focuses on the deep application of electronic case files. The common trend is that judicial artificial intelligence is moving from voice transcription, template documents, and simple retrieval toward a comprehensive application stage deeply integrated with electronic case files, workflow management, knowledge graphs, large language models, and trial supervision.
III. Institutional Value of Intelligent Assistance Systems First and foremost, the primary value of intelligent assistance systems lies in improving trial efficiency. Faced with a long-standing high volume of cases, judges must complete case file reviews, fact summarization, evidence examination, application of law, courtroom trial organization, and document drafting within a limited timeframe. Through material recognition, element extraction, summary of disputed issues, and assisted document generation, the intelligent system shoulders repetitive and administrative labor, enabling judges to devote more energy to factual determination, rule application, and value balancing. Shenzhen courts deconstructed the trial workflow into 85 nodes and assisted in a large volume of case filings and document generation during the pilot phase. The application of Suzhou’s “Future Judge’s Assistant” in intelligent case file review, document generation, and quality control error correction also demonstrates that artificial intelligence is reshaping traditional methods of reviewing files and producing documents.
Secondly, the deeper value of intelligent assistance systems lies in promoting the uniformity of adjudication standards. In the past, achieving similar judgments for similar cases primarily relied on judges’ proactive retrieval, ex-post case reviews, and guidance from higher courts. In contrast, intelligent assistance systems shift reference to similar cases, key rules, discretionary factors, and risk alerts forward into the case-handling process, enabling judges to obtain rule-based support before a judgment is formed. By transforming trial experience, adjudication rules, and case-handling essentials into a professional knowledge base, the Shanghai system essentially converts scattered, experiential, and tacit judicial knowledge into reusable, callable, and traceable rule resources.
Furthermore, intelligent assistance systems have altered the methods of case quality control. Traditional case evaluations are mostly retrospective, sample-based, and static, making it difficult to detect procedural flaws, factual omissions, and risks in the application of law in a timely manner. Through workflow node verification, document quality control, abnormal data identification, and risk early-warning, the intelligent system embeds supervision directly into the case-handling process. The large language model-based intelligent trial supervision of Beijing’s “Digital Intelligent Beijing Law” has been deployed to automatically screen for abnormal clues and alert users to risks in trial procedures and the application of law, reflecting a shift in trial supervision from ex-post error correction to process-oriented governance.
Moreover, for litigants, intelligent assistance systems lower the threshold for litigation through material prompts, procedural guidance, online interactions, and intelligent consultation. Following the integration of Zhejiang’s “Phoenix Intelligent Adjudication” with the financial comprehensive service platform, online channels for case filing, trial, and execution have been established. The Hangzhou Internet Court has explored full-process intelligence from case filing to the generation of judgment documents in specific cases such as financial loan contract disputes, breaking through the time and spatial constraints of traditional courtroom trials via an asynchronous trial mode.
Looking further, intelligent assistance systems have also enhanced judicial governance capabilities. When case elements, dispute types, industry risks, and litigation behaviors are structurally identified, courts can discover false litigation, batch disputes, hidden industry hazards, and weak links in governance from a massive volume of cases. Consequently, the intelligent trial assistance system is evolving from a case-handling support tool into an important digital infrastructure for the People’s Courts to participate in social governance.
IV. Risk Governance of Intelligent Auxiliary Trials The more deeply an intelligent assistance system penetrates trial workflows, the greater the need to simultaneously establish institutional constraints. The crux of judicial artificial intelligence lies in how these outputs are utilized, reviewed, and subjected to accountability.
First, algorithmic bias must be prevented. Judicial adjudication involves factual determination, admissibility of evidence, application of law, and value judgment. If training data contains historical biases, or if models develop inappropriate inclinations in similar case recommendations, risk identification, and document generation, the system may amplify existing issues. Therefore, normalized evaluations should be carried out for key functions such as factual element extraction, similar case recommendations, risk early-warning, and document generation, focusing on testing accuracy, stability, bias risks, false positives/negatives, and explainability. The processes of significant model updates and invocations in major cases must be tracked to ensure ex-post traceability and reviewability.
Second, automation bias and over-reliance must be prevented. As systems continuously provide summaries of disputed issues, adjudication paths, and preliminary document drafts, judges may tend to accept system conclusions under the pressure of efficiency, or even become subject to de facto algorithmic constraints. In response, the “auxiliary positioning” must be translated into explicit rules of authority: the system may flag risks, push similar cases, and summarize facts, but the final judgment on case outcomes must be completed independently by the judge. System outputs that exert a substantive impact on case disposition must undergo human review, with confirmations, corrections, or explanations for deviation recorded where necessary. The reasoning of a judgment must be independently articulated by the judge using legal language, and system conclusions must not replace judicial reasoning.
Third, the structure of liability must be clarified. Intelligent assistance systems are constructed with the joint participation of courts, technology enterprises, data governance teams, model training teams, and operations and maintenance (O&M) personnel. If a system erroneously extracts facts, omits evidence, mismatches similar cases, or generates faulty documents, liability cannot simply be attributed to the notion that “the judge is ultimately responsible”. While judges are responsible for adjudication conclusions, system design defects, data quality issues, and technical O&M risks should also be integrated into the scope of auditing, record-filing, and accountability. Courts must internally distinguish among judges’ review responsibilities, trial management responsibilities, technical O&M responsibilities, and suppliers’ security responsibilities, thereby avoiding situations where ambiguous liability leads judges to fear using the systems or where technical errors remain undetected.
In addition, data security and privacy protection must serve as fundamental constraints. Courts possess a vast amount of sensitive materials, including identity, property, commercial secrets, personal privacy, juvenile information, and criminal case records. Once case data is utilized for model training, cross-platform collaboration, and system O&M, risks of data leakage, unauthorized access, model inversion, and improper third-party contact inevitably rise. To address this, mechanisms for data classification, access control, desensitization processing, log tracking, and third-party security review should be established, safeguarding judicial security and the rights and interests of litigants through refined data governance.
The institutional goal of intelligent auxiliary trials is to preserve the subjectivity of judicial judgment while simultaneously enhancing efficiency, unifying adjudication, and strengthening supervision. Intelligent systems excel at discovering similarities, whereas judicial adjudication must identify differences. A mature intelligent trial assistance system should help judges determine where cases are identical, where they differ, and where explanations are required, rather than compressing the space for individualized justice through technical outputs.
Conclusion The boundaries of judicial artificial intelligence cannot be determined solely by technical capability itself, but must be jointly delineated by the operation laws of judicial power, procedural justice, and the ethics of responsibility. The standard for evaluating a whole-process intelligent trial assistance system should not merely be its technical advancement or the scale of its efficiency gains, but rather whether it achieves a higher degree of understandable, supervisable, and accountable judicial justice. Therefore, the ultimate goal of digital court construction is to render justice more prudent, transparent, and trustworthy under the support of technology.